November 2004
Volume 4, Issue 11
OSA Fall Vision Meeting Abstract  |   November 2004
Sparse coding for aerial images
Author Affiliations
  • Yoko Mizokami
    University of Nevada, Reno, USA
  • Michael A. Crognale
    University of Nevada, Reno, USA
Journal of Vision November 2004, Vol.4, 69. doi:
  • Views
  • Share
  • Tools
    • Alerts
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Yoko Mizokami, Michael A. Crognale; Sparse coding for aerial images. Journal of Vision 2004;4(11):69.

      Download citation file:

      © ARVO (1962-2015); The Authors (2016-present)

  • Supplements

Spatial receptive fields in primary visual cortex have been characterized as localized, oriented, and bandpass, comparable with basis functions derived from natural images (Olshausen & Field 1996, 1997). The properties of these fields may arise from the strategy of producing a sparse distribution of activity in response to these images. This suggests that sparse coding could characterize the properties of images in relation to responses in the cortex. Analyses have been based on image sampling from the “terrestrial” environment and not yet applied to the environment of aviators. We investigate whether characteristics of images in the aerial environment differ from those of ground-based images and how differences might affect visibility.

We compared the characteristics of images of aerial and ground environments using the sparse coding technique. The training process included thirty 512 × 512 pixel images deriving two hundred basis functions from 12 × 12 pixel image patches. Overall, the basis functions learned from aerial images were noisier than those from ground images suggesting that the characteristics of aerial images differ from those of terrestrial scenes. The distributions of orientation of the functions were more vertical and horizontal, and those of spatial frequency had clearer peaks in ground images.

In order to further characterize the response of the cortex to the aerial environment we applied basis functions learned from terrestrial images to both land and aerial-based images and compared the weightings (coefficients) of the basis functions for these two classes. Results indicate larger variability (poorer fit) of terrestrial functions to aerial images.

The variety of results from aerial images suggests that our visual system which is adapted to the terrestrial environment may not be optimized to the aerial environment. This may affect visibility in the aerial environment.

Mizokami, Y., Crognale, M. A.(2004). Sparse coding for aerial images [Abstract]. Journal of Vision, 4( 11): 69, 69a,, doi:10.1167/4.11.69. [CrossRef]
 Supported by a grant from the Department of Transportation, the Federal Aviation Administration to M. Crognale

This PDF is available to Subscribers Only

Sign in or purchase a subscription to access this content. ×

You must be signed into an individual account to use this feature.